What does current economic research indicate about the potential for AI to displace jobs in the next decade?

Current economic research on AI and job displacement paints a nuanced picture: while some projections warn of massive job losses, the actual data so far shows very little net displacement, and many jobs are being reshaped rather than eliminated. Here’s what the latest studies indicate for the decade ahead.

Early signals: what we’ve seen so far

• The best early evidence finds limited signs that AI has reduced overall employment. No systematic rise in unemployment has appeared among highly exposed workers since late 2022, though hiring of younger workers has slowed in those occupations [1][2][67].
• By December 2023, AI had not caused major changes in total US employment—losses in heavily exposed roles were mostly offset by gains elsewhere and by hiring at firms that adopted AI [50][51].
• Looking back further, from 2014 to 2023, roles heavily exposed to AI did not lose ground relative to other jobs, thanks to offsetting forces [56].
• More than 33 months after ChatGPT’s launch, the broader labor market still shows no discernible disruption, and current AI‑exposure metrics have no meaningful relationship with employment or unemployment changes [62][64].
• AI is nowhere near its theoretical potential—actual workplace usage is only a fraction of what is technically feasible [5][69].
• That said, some displacement is already appearing. About 5.1% of US jobs (7.9 million) currently face high automation‑displacement risk, and occupations with higher risk shares have experienced sharper declines in labor demand since late 2022 [14][15][17]. In May 2023, 3,900 US job losses were directly linked to AI [72], and 23.5% of US companies report having replaced some workers with ChatGPT or similar tools [40].
• Worker anxiety is high: 30% of US workers fear their job will be replaced by AI or similar technology by 2029 [77].

Where projections point for the next decade

• The U.S. Bureau of Labor Statistics now explicitly includes AI impacts in its 2023–33 employment projections [78]. Occupations with higher observed AI exposure are projected to grow more slowly through 2034: each additional 10‑percentage‑point of AI task coverage is associated with a 0.6‑percentage‑point drop in projected growth [3][4][28][29][68].
• Specific occupations forecast to shrink by 2033 include bank tellers (–15%), customer service representatives (–5%), medical transcriptionists (–4.7%), and credit analysts (–3.9%) [34][35][36][37]. Meanwhile, software‑developer employment is expected to jump 17.9% over the same period [45].
• More sweeping estimates circulate: about 30% of current US jobs could be automated by 2030 [38][70], and globally up to 300 million jobs (9.1%) could be lost to AI [39][71]. By 2030, 14% of all employees worldwide may be forced to change careers because of AI [41][75]. In advanced economies, nearly 60% of jobs could be impacted in some way [43][76].
• For the very near term (2–3 years), one analysis suggests that 50% to 55% of US jobs will be reshaped by AI—tasks change, but workers keep their jobs [7][33].

The role of adaptive capacity

• Exposure alone doesn’t mean a worker is doomed; their ability to adapt makes a huge difference. Roughly 26.5 million highly AI‑exposed workers (70% of the exposed group) have strong adaptive capacity—good savings, transferable skills, geographic mobility, and favorable age—so they can navigate job transitions [9][10]. This capacity is often missing from older exposure measures [13].
• On the other hand, about 6.1 million workers—overwhelmingly in clerical and administrative roles, 86% women—lack that cushion. They tend to have limited savings, narrow skill sets, advanced age, or live in areas with few job alternatives [8].
• Geographically, these vulnerable high‑exposure jobs are concentrated in college towns and state capitals, especially in the Mountain West and Midwest [12].
• Researchers are now building frameworks specifically to identify at‑risk jobs before displacement becomes visible, which should help policymakers and businesses prepare [6].

Which jobs and workers face the biggest risk?

• Routine, lower‑skill roles remain the most threatened: manufacturing, retail, customer service, and transportation top the list [18][19][22]. Sector‑specific estimates suggest potential displacement of 30% in manufacturing, 25% in customer service, 20% in retail, 15% in transportation, and about 10% in agriculture (though at a slower pace) [23][24][25][26][27].
• Entry‑level jobs are especially vulnerable, with nearly 50 million US jobs at risk in the coming years [42].
• At the same time, AI tends to expand high‑skill roles. Over five years, high‑wage roles heavily exposed to AI saw their share of total employment grow by about 3% [54]. Legal jobs, for instance, are predicted to increase 6.4% in employment thanks to AI [58].
• The result is employment polarization: high‑ and low‑wage jobs expand while middle‑skill roles hollow out [20].

Net impact: job killer, job expander, or both?

• At the firm level, AI adoption is linked to higher headcount growth; one analysis of over a billion job ads suggests AI may be a job expander when companies use it to enter new markets and improve products [30][31][50].
• However, the effect depends on a job’s task structure. When AI can handle most tasks in a role, that role’s share within a firm falls by about 14%. But when AI touches only a few tasks, employment in that role can actually grow [52][53].
• Some occupations that closely match AI’s current abilities have already shrunk: business, financial, architecture, and engineering roles declined by roughly 2% to 2.5% over five years [57].
• Indirect effects also appear. Food‑service employment has declined not because AI can cook, but because employers that don’t adopt AI grow more slowly, dragging down demand for all their workers [59].
• Academics debate whether AI will complement human labor or create enough new jobs to offset displacement, especially if skill development lags [21].
• Crucially, these patterns come from the pre‑generative‑AI era. It remains an open question whether tomorrow’s generative AI will amplify displacement—by gobbling up the tasks humans currently shift to—or lift all boats through broad productivity gains [60][61][63].

Lessons from history and the big picture

• The OECD finds “little evidence so far that AI is leading to job losses” [47], even though occupations at highest automation risk make up 27% of employment across its member countries [48]. AI’s impact is likely to be both positive and negative—automating repetitive tasks while also creating new challenges [46].
• At the same time, AI could enhance human work by boosting productivity, spurring growth, and creating fresh opportunities [49].
• Historically, major technological disruptions play out over decades, not a few years, so any widespread job effects from AI are likely to take longer than the 33 months we have observed so far [65][66]. Some forecasts place AI’s most disruptive years in the 10‑ to 30‑year window [44].
• On the wage side, automation and AI can displace workers from tasks where they had a comparative advantage, lowering wages and employment for exposed groups [79][80]. AI‑driven automation may also reduce labour demand in some areas, putting downward pressure on wages [81].

Bottom line

Current economic research does not suggest that AI will cause mass joblessness in the next decade, but it does foresee considerable reshaping. Occupations with high AI exposure will likely grow more slowly, some routine roles will decline, and vulnerable workers with limited adaptive capacity face real displacement. However, overall employment may hold steady or even expand as AI boosts productivity and creates new roles—especially for workers who can adapt. The biggest unknown is how fast generative AI evolves and how quickly workers and policies adjust.